Body mass index and risk of malnutrition in community-living elderly men and women: relationships with morbidity, mortality and health- related quality of life. The Tromsø and HUNT studies

(1)

FACULTY OF HEALTH SCIENCES

DEPARTMENT OF COMMUNITY MEDICINE

Body Mass Index and Risk of Malnutrition in Community-living Elderly Men and Women:

Relationships with Morbidity, Mortality and Health-related Quality of Life.

The Tromsø and HUNT studies

Department of Community Medicine

University of Tromsø

Jan-Magnus Kvamme

A dissertation for the degree of

Philosophiae Doctor

(2)

(3)

Preface: From clinical practice to epidemiological research

During my more than 20 years in clinical medicine and gastroenterology, I have met numerous elderly patients with weight loss and who were underweight. For some patients, these problems were symptomatic of a severe disease, such as cancer. For others, however, malnutrition was the missing puzzle piece in a complex case history. In the hospital system, malnutrition and

underweight had often gone unrecognised. This fact wakened my interest in malnutrition and Professor Jon Florholmen introduced me to this field.

Most of my elderly patients were from the community of Tromsø, and the Tromsø Population Study had for years explored the health of its inhabitants. The Tromsø 6 survey was in the planning phase when I prepared my PhD project. It is a short distance from the University Hospital in Tromsø to the epidemiological setting at the Department of Community Medicine, where I met Professor Bjarne K Jacobsen. He has a background both in nutritional science and epidemiology. During these last three years, I have had the privilege to more carefully investigate the interesting relationship between nutritional-status and health in the elderly population.

Jan-Magnus Kvamme Tromsø, March 2011

(6)

Acknowledgements

This research was done at the Department of Community Medicine, University of Tromsø, between 2008 and 2011 and was funded by a PhD grant from The Northern Norway Regional Health Authority, Medical Research Programme (Helse Nord, Senter for Aldersforskning).

I would particularly want to thank the following persons for their valuable help:

My principal advisor, Bjarne K Jacobsen, for his friendly and critical guidance, his patience and for sharing his significant knowledge concerning aspects of nutrition, statistical methods and epidemiology.

Jon Florholmen, my co-supervisor, for his clinical contributions and consistently optimistic attitude.

Tom Wilsgaard for both brief and more extensive discussions regarding statistical problems.

Jan Abel Olsen for important suggestions regarding the work dealing with quality of life aspects, and Ole Grønli for sound advices regarding the SCL-10 analyses.

My other co-authors from HUNT, Jostein Holmen and Kristian Midthjell for a good collaboration.

My colleague, Eyvind Paulssen, for encouraging conversations regarding research and working on a thesis, and my other colleagues at the Department of Gastroenterology for taking on a larger share of work during my period of absence.

My fellow PhD candidate, Bente Morseth, my colleagues and the staff at the Department of Community Medicine in addition to Torgeir Engstad and Pål Stenumgård, for valuable discussions.

The HUNT Study (Helseundersøkelsen i Nord Trøndelag) for access to data used in the mortality analyses.

My parents, Astrid and Jan-Ivar, for their interest and commitment. My wife, Brita, for continuous support during the ups and downs throughout the project, and our three children Rasmus, Ragnhild and Ingeborg, for reminding me of the other important aspects of life in this busy period.

(7)

Kort norsk sammendrag-Short Norwegian summary

I denne avhandlingen er forholdet mellom ernæringsstatus og sykelighet, dødelighet og helse- relatert livskvalitet undersøkt hos hjemmeboende eldre. Avhandlingen er basert på de tre befolkningsundersøkelsene, Tromsø 4 (1994-95), Tromsø 6 (2007-08) og Helseundersøkelsen i Nord-Trøndelag, HUNT 2 (1995-97). Ernæringsstatus er vurdert både ved hjelp av

kroppsmasseindeks (KMI, kg/m²), også kjent som BMI, og et spesielt screeningsverktøy for underernæring, MUST (Malnutrition Universal Screening Tool).

Vurdert utifra MUST fant vi at 8 % av hjemmeboende eldre hadde middels eller høy risiko for underernæring. Både depressive symptomer, gjennomgått brudd (lårhals), redusert muskelstyrke og dagligrøyking forekom hyppigere hos undervektige (KMI <20 kg/m² ). Redusert fysisk aktivitet og kronisk lungesykdom var vanligere både ved undervekt og fedme (KMI ≥30 kg/m²) (U-formet relasjon til KMI). Eldre med fedme hadde også økt hyppighet av diabetes og iskemisk hjertesykdom. Helse relatert livskvalitet var redusert ved økende risiko for underernæring, mer hos menn enn hos kvinner. Dødeligheten var økt hos alle med KMI under 25 kg/m² og en del av dette kunne forklares av dødelighet på grunn av lungesykdommer. Den laveste dødeligheten fant vi hos de som var overvektige (KMI 25-30 kg/m² hos menn og 25-32.5 kg/m² hos kvinner). Det var moderat økt dødelighet hos eldre med fedme.

Underernæring utgjør altså en betydelig helserisiko for hjemmeboende eldre og det er viktig å få på plass tiltak som gjør at underernæring oppdages i en tidlig fase. Andre studier har vist at helsetilstanden kan bedres hvis underernæring behandles.

(8)

English summary

The elderly population is rapidly growing, and elderly individuals are more vulnerable to nutritional problems than other adults. A number of studies have found evidence for adverse health outcomes in elderly patients at risk of malnutrition. However, much of the previous research has been performed in hospital populations or in groups of elderly patients with specific diagnoses. More population-based studies in this area are therefore needed. Increased

understanding of nutrition in elderly individuals can contribute to the identification of individuals at risk of malnutrition at an earlier stage.

Based on data from three large health surveys, we aimed to study the relationship between nutritional status and important health outcomes in community-living elderly individuals. We first explored the associations between disease burden, social and life style variables in a cross- sectional design based on data from the Tromsø 4 survey (1994-1995) (paper I). We found that fractures (hip), mental distress, reduced muscle strength and current smoking were more prevalent in underweight individuals. Chronic lung disease and reduced physical activity had a U-shaped relation with body mass index (BMI). Diabetes and ischemic heart disease were more prevalent in obese individuals.

In the Tromsø 6 survey (2007-08), we included the Malnutrition Universal Screening Tool (MUST). We found that approximately 8% of this community-living population was at medium or high risk of malnutrition and that 20% was obese (BMI ≥30 kg/m²). We used a cross-sectional design to explore the association between mental health symptoms and both risk of malnutrition and BMI (paper II). Mental health symptoms were significantly associated with the risk of

(9)

malnutrition, and an association was also found for subthreshold mental health symptoms.

Quality of Life is an important health aspect for the increasing number of elderly individuals with longer life expectancy. We found a significant reduction in Health-related Quality of Life with an increasing risk of malnutrition, and this was more pronounced in men than in women (paper III).

In paper IV, we combined the HUNT 2 (The Nord-Trøndelag Health Study, 1995-97) and Tromsø 4 surveys with the intention of exploring the relationship between BMI categories and both total and cause-specific mortality in a prospective design. We found mortality to be increased in all BMI categories below 25 kg/m² and that overweight individuals had the lowest mortality (BMI 25-29.9 kg/m² in men and 25-32.4 kg/m² in women). A modest increase in mortality was found with increasing BMI among obese men and women. About 40% of the excess mortality in the lower BMI range in men was explained by mortality from respiratory diseases.

This thesis describes increased morbidity, mortality and reduced HRQoL in community-living elderly individuals at risk of malnutrition or with lower BMI. These findings emphasise the importance of nutritional screening, especially in primary care. Previous research has

demonstrated that nutritional intervention can reduce adverse health outcomes in elderly at risk of malnutrition.

(10)

List of papers

This thesis is based on the following four papers, which are referred to in the text by their Roman numerals:

Paper I

Kvamme J-M, Wilsgaard T, Florholmen J, Jacobsen B K. Body mass index and disease burden in elderly men and women: The Tromsø Study.

European Journal of Epidemiology. 2010 Mar; 25 (3):183-93. Epub 2010 Jan 20.

Paper II

Kvamme J-M, Grønli O, Florholmen J, Jacobsen B K. Risk of malnutrition and mental health symptoms in community living elderly men and women: The Tromsø Study.

Submitted.

Paper III

Kvamme J-M, Olsen J A, Florholmen J, Jacobsen B K. Risk of malnutrition and Health related Quality of life in community-living elderly men and women: The Tromsø Study.

Quality of Life Research. 2011 May;20(4):575-82. Epub 2010 Nov 13.

Paper IV

Kvamme J-M, Holmen J, Wilsgaard T, Florholmen J, Midthjell K, Jacobsen B K. Body mass index and mortality in elderly men and women: The Tromsø and HUNT studies.

Journal of Epidemiology & Community Health. 2011 Feb 14. (Epub ahead of print)

(11)

Abbreviations

BAPEN - British Society of Parenteral and Enteral Nutrition BMI - body mass index

CI - confidence interval

CONOR-COhortNORway, a collaboration between several health surveys in Norway COPD - chronic obstructive pulmonary disease

CVD - cardiovascular diseases

DXA - dual-energy X-ray absorptiometry

EQ-5D index - value attached to an EQ-5D state according to a particular set of weights EQ VAS - standard vertical visual analogue scale

ESPEN- The European Society for Clinical Nutrition and Metabolism (previously European Society of Parenteral and Enteral Nutrition)

GDS - Geriatric Depression Scale HR - Hazard ratio

HRQoL - Health-related Quality of life

HUNT - The Health Study of Nord-Trøndelag County, Norway ICD - International Classification of Diseases

IHD - ischemic heart disease MNA - Mini nutritional assessment

MUST - Malnutrition Universal Screening Tool NRS 2002 - Nutritional risk screening 2002 OR - odds ratio

PEM – protein energy malnutrition

SCL-10 - (Hopkins) Symptoms check list-10 S.D. - standard deviation

SPSS - Statistical package for the Social Sciences WC - waist circumference

WHO - World Health Organization

(12)

Definitions

More details can be found in the respective sections indicated by the page number(s).

BMI: body mass index, weight divided by height squared (kg/m²), page 28 and 47.

Community Living: Present place of residence is a private dwelling as distinct from elderly hospital in-patients or nursing home residents, page 14 and 43.

Older person: individual aged ≥65 years [145, 154], page 15.

Health related Quality of Life: The measurable impact of a person’s perception of his or her health and the effect that produces on satisfaction with life and well-being [154], page 19.

Morbidity: from Latin morbidus, state of being diseased [154]

Malnutrition: A state in which a deficiency, excess or imbalance of energy, protein and other nutrients causes adverse effects on body form, function and clinical outcome [143], page 14.

Underweight: BMI ≤20 kg /m²[143], page 61.

Overweight: BMI 25-29.9 kg/m²[164], page 48.

Obesity: BMI ≥30 kg/m²[164], page 48.

Sarcopenia: from Greek sarx for flesh and penia for loss. Loss of muscle mass, muscle endurance and muscle force occurring during aging [118], page 62.

(13)

List of tables and figures

Figure 1 Location of Tromsø Municipality, page 23.

Figure 2 Location of North Trøndelag County, page 25.

Figure 3 The Malnutrition Universal screening Tool, page 29.

Figure 4 Timeline for the follow-up in paper IV, page 31.

Figure 5 The aspects of health described in papers I through IV, page 37.

Figure 6 Internal and external validity with examples from the thesis, page 44.

Figure 7 Sarcopenia as a common mechanism for several of the observed relations, page 63.

Table 1 Studies on the relationship between depression/mental health and the risk of malnutrition; non-hospitalised populations, page 17.

Table 2 Selected studies on the relationship between BMI and mortality, page 20.

Table 3 Study participants included in the studies referred to in this thesis (I-IV), page 23.

Table 4 Number of invited, attending and included men and women in the Tromsø 4 survey, page 26.

Table 5 Number of invited, attending and included men and women in the Tromsø 6 survey, page 26.

Table 6 Number of invited and attending participants in The Tromsø 4 and HUNT 2 surveys combined, page 27.

(14)

1 Introduction

1.1 General Introduction

The elderly population is rapidly growing, and elderly people live longer than ever before. By 2050 it is expected that one in three Europeans will be 60 years of age or older [153]. In Norway, approximately one in seven individuals is 65 years of age or older. The large majority of the elderly population is self-reliant and live in their homes (i.e., community-living). In the city of Tromsø, nursing home residents constitute only about 6% of the population in this age category [24], similar to the Norwegian population as a whole [137]. However, a large portion of health care resources are allocated to the elderly population, and one in three hospital beds is occupied by individuals in this age category [126].

Health is closely related to nutrition, and in comparison to younger adults, elderly individuals are more vulnerable to nutritional problems. There is a large body of evidence supporting the idea that elderly people are at risk for malnutrition [143]. However, many of the studies available have been performed in hospitalised patients and in groups of elderly individuals with specific diseases or conditions. Thus, more population-based research is needed in the field of nutritional

problems in elderly people, with special attention to the lower BMI categories and elderly individuals at risk for malnutrition. This is the main focus of the present thesis.

There is no universally accepted definition for the term malnutrition [143]. The word can be literally translated as “bad nutrition”. The following definition has been suggested by Professor M. Elia: “Malnutrition is defined as a state in which a deficiency, excess or imbalance of energy, protein and other nutrients causes adverse effects on body form, function and clinical outcome.”

(15)

[95, 143]. Strictly speaking, this also includes obesity, but, as in most studies of elderly individuals, excess weight is not included in the concept of “risk of malnutrition” used in this thesis. Protein energy malnutrition (PEM) results if an individual’s needs for protein and energy are not satisfied by the diet [149], and this corresponds to the malnutrition term used in this thesis. The nutritional screening tool used in two of the included papers was developed to detect PEM [140]. PEM as a term is about to be replaced by an approach to malnutrition that

incorporates inflammatory response [75] (see section 7.3, page 67 for more details).

Older persons are often defined as individuals aged ≥65 years [145, 154] and this age limit is applied here to define the elderly population.

1.2 Aging and Nutritional status

Concurrent with the epidemic of obesity, malnutrition also seems to be a persistent problem in the affluent parts of the world and is more prevalent in elderly individuals than in other adults [111]. In developed countries, malnutrition is largely related to diseases [143].

A range of physiological and biological changes involved in ageing affect nutritional status and combine to make deficiencies of macro- and micronutrients more common in elderly individuals.

Taste and smell, two senses important for the pleasure of eating, are often decreased during aging [65]. This sensory deficit may partly explain reduced appetite, which can, in turn, contribute to malnutrition [101]. Compared to younger adults, elderly individuals have less ability to respond to concurrent underfeeding that may occur in concert with acute diseases [117]. Gut mechanisms, including a delay in gastric emptying that result in early satiety, may also contribute to lowered

(16)

food intake in elderly individuals [156, 160]. Approximately the same requirements of vitamins and micronutrients should be ingested in a context of reduced food intake [161], a fact that may also contribute to increased risk of micronutrient deficiencies in the elderly.

Generally, body weight and BMI increase during life up to the age of 50-60 years, after which these values level off and slowly decrease in old age [60, 156]. Body composition also changes with advancing age. The relative proportion of adipose tissue is increased and muscle tissue mass, quality, and strength is reduced, with the latter often characterised as sarcopenia [156]

[118]. There are a number of risk factors for sarcopenia, including malnutrition, low protein intake, several chronic health conditions and the ageing process itself [26]. (See section 7.3, page 62 for further details).

1.3 Morbidity and Nutritional status

Morbidity is increased in the elderly, and it was found in a previous study that 82% of elderly people had one or two chronic health problems and that 65% had two and more chronic health problems [167]. Nutritional status may be closely related to chronic health problems. Most previous studies of associations between BMI and various medical conditions have focused either on the detrimental effect of on obesity [19, 109], adult populations without specific analysis of elderly participants, [3, 110] or selected chronic diseases [20, 56]. A number of studies of primarily hospital populations have found malnutrition to be a common problem in elderly patients with severe or chronic diseases [143]. However, there are fewer population-based studies that examine elderly persons in all BMI categories. Thus, important factors associated with low BMI compared to other BMI categories may not have been identified.

(17)

1.4 Mental health and risk of malnutrition

Mental health problems contribute to the increased morbidity in elderly people. Anxiety and depression, often seen as co-morbid conditions with overlapping symptoms [97], are the two most frequently occurring mental health disorders [31]. Malnutrition is also relatively common in elderly individuals and may be associated with mental health, particularly depression [12].

Although several studies have found mental disorders to be a risk factor for involuntary weight loss and malnutrition in geriatric inpatients and outpatients [159], little population-based research has been done on the relationship between the risk of malnutrition and mental health in the elderly.

The relationship between malnutrition and depression has been assessed in some previous studies by the Geriatric Depression Scale (GDS) and the Mini Nutritional Assessment (MNA) instrument (table 1).

Authors Participants Methods and variables Results Johansson et

al. 2009 [79]

579 community- living elderly (Sweden)

Prospective study MNA + GDS

Depressive symptoms predictive of malnutrition Smoliner et

al. 2009 [130]

114 nursing home residents

(Germany)

Cross-sectional study MNA + GDS

Modest association between depressive symptoms and malnutrition

Cabrera et al.

2007 [18]

267 community- living elderly individuals (Brazil)

Cross-sectional study, MNA and GDS/regular use of antidepressant

medicines

Association between depressive symptoms and malnutrition

Abbreviations: MNA, Mini Nutritional Assessment (nutritional screening tool), GDS, Geriatric Depression scale

Table 1 Studies on the relationship between depression/mental health and the risk of malnutrition; non-hospitalised populations.

(18)

Johansson et al. found in a Swedish study of 579 community-living elderly individuals that depressive symptoms were predictive of malnutrition [79], a finding that was observed to a larger extent in men than in women. Smoliner et al. examined nursing home residents and found no differences in the mean MNA score between subjects who had depression and those who did not.

However, a modest association was demonstrated between malnutrition and depression in a regression analysis [130]. A study of 267 community-living elderly individuals in Brazil [18]

showed a positive relationship between malnutrition and depression.

The relationship between BMI and depressive symptoms in adults and elderly individuals has in previous studies been assessed with a main focus on higher BMI categories. Limited evidence for an association between obesity and depressive symptoms has been found [5]. Studies restricted to elderly individuals have yielded conflicting results. In one study, obese elderly men were found to have a reduced risk of depression [107], whereas a later study reported an increased risk of depression in obese individuals [125].

Thus, larger studies examining the associations between BMI, the risk of malnutrition and mental health in community-living elderly individuals are needed.

1.5 Health-related quality of life and risk of malnutrition

Quality of Life (QOL) has received increasing attention in recent decades as a measure for comparing the health status of different patient groups and for measuring health outcomes. QOL is not a well-defined term. However, many will argue that most people, at least in the developed world, are familiar with the term and have an intuitive understanding of it [43]. Aristotle (384-

(19)

322 BC) described a term (eudemonia) regarded as very close to the modern QOL in his work on ethics (Nicomachean Ethics) [43]. He stated that the meaning of the term varies from person to person and depends on the life situation of each individual. To differentiate between QOL in a general sense and the need for a more precise definition in clinical medicine, the narrower term, Health Related Quality of Life (HRQoL) is often used. However, this terminology is also not very precisely defined. In Taber’s Medical Dictionary (2009) [154], HRQoL is defined as: “The measurable impact of a person’s perception of his or her health and the effect that produces on satisfaction with life and well-being.” This corresponds to the HRQoL term used in this thesis.

Despite the previously mentioned evidence of increased morbidity [88, 143] and mortality (see below) [63, 142] in elderly people at risk of malnutrition, little attention has been given to the ways in which malnutrition affects HRQoL. For the increasing number of elderly individuals with longer life expectancies, not only is the duration of life important, but also the quality of those additional years.

The concept of HRQoL broadens a previous definition of health based on morbidity and

mortality to include aspects of health that include subjective assessments of physical, emotional and social functioning [33]. Nutrition may affect both the physical and psychological aspects important for HRQoL [143]. Several reports have found HRQoL to be reduced in obese

individuals [77]. In a study of nursing home patients [25] and a smaller, community-based study [81], QOL was reduced in elderly patients at risk of malnutrition.

However, larger, community-based studies evaluating HRQoL in elderly individuals at risk of malnutrition are lacking.

(20)

1.6 Body mass index and mortality

The impact of BMI on mortality in the elderly population is still controversial. There is a growing concern regarding the increase in mortality related to obesity [105], whereas mortality associated with underweight has gained less attention.

As reviewed by Heiat (2001) [62], Zamboni (2005) [171], and Janssen (2007) [73], the recommendations of ideal weight for adults [164] seem to be too restrictive for elderly individuals, in whom being moderately overweight is of limited risk with respect to increased mortality. It is not possible in this introduction to present the more than fifty studies included in these review articles, in addition to other BMI-mortality studies. However, some selected studies are presented in table 2 with the intention to include studies also from Scandinavian countries.

Authors Participants Design Results

Total mortality

Results

Cause specific mortality

Comments

Waaler 1984 [158]

1 800 000 from Norway

Prospective study based on

compulsory X-ray

examinations for

tuberculosis 1963-65

Marked U- shape between BMI and mortality, lowest mortality 21-25 kg/m², somewhat higher in individuals aged

≥65 years

In both men and women 50-64 years of age, highest mortality in the low BMI categories from obstructive lung diseases, tuberculosis, lung and stomach cancer

Cause specific mortality not examined in persons ≥65 years

Dey et al 2001 [29]

2 628 men and women aged 70 years from Sweden

Prospective study, up to 15 years follow up, BMI (quintiles)

Lowest BMI quintile, highest mortality.

Lowest mortality in BMI 27-29 kg/m² in males and 25-27 kg/m² in women (non smoking)

Not explored

(21)

Rissanen 1991 [116]

17 159 women aged 25-79 years (1 437 women aged

≥65 years) from Finland

Prospective, 12 years follow-up

In women aged

≥65 years, mortality varied little with BMI

Thinness seemed to predict deaths from cancer

Deaths from respiratory diseases not explored

Engeland 2003 [38]

2 million men and women, The majority included in 1963-75

Prospective 22 years follow-up

Subgroup analyses, elderly of 70-74 years: in men optimal BMI 24 kg/m², and in women, optimal BMI: 25.7 kg/m²

Not explored Partly extension of Waaler 1984.

Smoking information available for a sub-cohort.

Increased risk at low BMI also in non-smokers.

Pischon 2008 [112]

359 387 adults from nine

countries in Europe

Prospective 9.7 years follow up

Subgroup analysis of ≥65 years, lowest mortality men BMI 25-26.6 kg/m² and women 23.5-25 kg/m².

Significant increased mortality if BMI <21.0 kg/m² (men and women)

Explored in adults, not specifically in elderly: Respiratory disease mortality increased if BMI

<23.5 kg/m²

Several analyses also adjusted for waist hip data.

Table 2 Selected studies on the relationship between BMI and mortality (only studies with measured height and weight). Results reported with primary attention towards the lower BMI categories.

In Norway, the relationship between BMI and mortality was examined early by Waaler (Waaler 1984) based on height and weight data obtained during screening for tuberculosis and with deaths in the period from 1963-1979 [158]. Waaler described a U-shaped relationship between BMI and

(22)

mortality. In adults, the strongest increase in mortality in the lower BMI categories was found for obstructive lung disease. However, BMI- mortality curves for cause-specific mortality in elderly individuals were not examined. Two large studies with primarily adult participants from several different countries were published in 2008 and 2009. In the first of these studies, the lowest mortality rate was found to be at a BMI of approximately 22.5-25 kg/m² for all ages [163]. The second of these studies (Pischon 2008) found the lowest mortality to be at a slightly higher BMI levels (table 2) [112]. Both studies provide several subgroup analyses, but very limited analysis of elderly participants. Dey et al. (Dey 2001) [29] found that overweight elderly, individuals had the lowest mortality, a finding also demonstrated by others [73].

To summarise, a number of studies have found increased mortality in underweight individuals, but this association remains to be fully explained in the elderly population. Thus, more research should be done on the relationship between BMI and mortality (total and cause specific) with special attention to the lower weight categories.

2 Aims of the thesis

There is a need for more population-based research on the relationships between nutritional status and important health outcomes, including morbidity, HRQoL and mortality in elderly people, with special attention to malnutrition and underweight individuals. More specifically, this thesis concentrates on the following questions regarding community living elderly people:

1. How common is the risk of malnutrition and what are the prevalences of the different BMI categories?

2. What are the characteristics and disease burden of elderly people in different BMI categories, specifically the lower BMI categories?

(23)

3. What is the association between mental health symptoms and the risk of malnutrition?

4. Is there an association between the risk of malnutrition and impaired HRQoL?

5. What is the relationship between BMI and both total and cause-specific mortality?

3 Subjects

The individuals included in these analyses were participants in the 4^th and 6^th surveys of the Tromsø Study (Tromsø 4 and Tromsø 6) and the 2^nd Nord Trøndelag Health survey (HUNT 2) (see table 3).

Tromsø 4 1994-95

HUNT 2 1995-97

Tromsø 6 2007-2008 Paper I n = 4 259

Paper II n = 3 111

Paper III n = 3 286

Paper IV n = 16 711

Table 3 Study participants included in the studies referred to in this thesis (Paper I-IV).

3.1 The Tromsø 4 survey (paper I and IV)

The Tromsø Study is a single-centre, population-based longitudinal study with repeated health surveys of the Tromsø municipality [72]. Tromsø is the largest city in the northern part of Norway, with a population of 68 000. Most inhabitants live on the Tromsø Island (figure 1).

Figure 1 Tromsø Municipality

(24)

The fourth cross-sectional health survey of the Tromsø population was conducted in 1994-1995.

All community-living inhabitants in the municipality aged 25 years and older were invited to participate, but the present analyses are restricted to participants aged 65 years and older. Nursing home residents were invited, but very few participated. A total of 5 892 subjects in this age group were invited and the study sample consisted of 4 259 persons (72% of the invited). One reminder letter was sent. The participation rate declined with age (table 4). In subjects aged 80 years and older, less than 50% participated.

3.2 The HUNT 2 survey (paper IV)

The Nord Trøndelag Health study (HUNT study) is a large, population-based longitudinal study with repeated health surveys of the county of Nord-Trøndelag. This is a sparsely populated and largely rural county located in the central part of Norway, with 127 000 residents (figure 2). The second Nord-Trøndelag health survey (HUNT 2) was conducted in 1995-97 and has been previously described in more detail [68]. All community-living inhabitants aged 20 years and older were invited to participate in the survey, but only participants aged 65 years and older were included in the study population described in paper IV. A total of 21 946 individuals in this age group were invited and 15 250 participated in the study. One reminder letter was sent. The overall participation rate was rate was 70% and declined with age. In subjects aged 79 years and older, 46% participated. Nursing home residents were invited, but very few participated (45 out of the 15 250 participants were permanent nursing home residents).

(25)

Data from the HUNT 2 and the Tromsø 4 surveys were combined for the analyses presented in paper IV. Details about participation by increasing age in the combined cohort are found in table 6. These two surveys had been pre-planned to include the same core questions. Mean BMI, smoking habits and household income in the two regions Nord-Trøndelag and Troms were quite similar [136, 168]. The inhabitants of Tromsø constitute approximately half of the population of the Troms County.

Figure 2 Nord Trøndelag County

3.3 The Tromsø 6 survey (paper II and III)

The Tromsø 6 survey was conducted between October 2007 and December 2008. Invitations were sent to all community-living inhabitants aged 25 to 87 years. One reminder letter was sent.

The analyses in this thesis are restricted to individuals aged 65 years and older. A total of 6 098 subjects in this age category were invited and 4 017 subjects (66%) participated in the study.

Nursing home residents were also invited, but very few participated (only 8 of the 4 017 participants were permanent nursing home residents). The age distribution of the groups in the invited population, the attending population and the two study samples for papers II and III, respectively, are described in table 5. The two study samples consist of relatively few of the eldest participants, and relatively more women than men were excluded because of missing values.

(26)

Age range

Invited, Tromsø 4 N

Attending, Tromsø 4 n (% of invited)

Paper 1 Study sample n (% of invited)

(years) Men Women Men Women Men Women

65–69 810 970 691 (85.3) 860 (88.7) 683 (84.3) 841 (86.7) 70–79 1 216 1 548 935 (76.9) 1 240 (80.1) 921 (75.7) 1 215 (78.5)

80– 414 934 214 (51.7) 411 (44.0) 208 (50.2) 391 (41.9)

All 2 440 3 452 1 840 (75.4) 2 511 (72.7) 1 812 (74.3) 2 447 (70.9)

Table 4 Number of invited, attending and included men and women in the Tromsø 4 survey. In the study sample, participants not willing to take part in the research or with missing values for height and/or weight were excluded.

Age range

Invited, Tromsø 6 N

Attending, Tromsø 6 n (% of invited)

Paper ll Study sample, n (% of invited) SCL10

Paper lll Study sample, n (% of invited) HRQoL

(years) Men Women Men Women Men Women Men Women

65–69 1 068 1 054 830 (77.7) 827 (78.4) 721 (67.5) 662 (62.8) 741 (69.4) 690 (65.5) 70–79 1 197 1 456 841 (70.3) 988 (67.9) 698 (58.3) 699 (48.0) 732 (61.2) 748 (51.4) 80–87 492 831 196 (39.8) 335 (40.3) 139 (28.3) 192 (23.1) 159 (32.3) 216 (26.0) All 2 757 3 341 1 867 (67.8) 2 150 (64.4) 1 558 (56.5) 1 553 (46.5) 1 632 (59.2) 1 654 (49.5)

Table 5 Number of invited, attending and included men and women in the Tromsø 6 survey. In the study samples, participants with

(27)

Age range

Invited, Tromsø 4 and HUNT-2

N

Attending, Tromsø 4 and HUNT-2

n (% of invited)

Paper IV Study sample

n (% of invited)

(years) Men Women Men Women Men Women

65–69 3 478 3 786 2 898 (83.3) 3 231 (85.3) 2 635 (75.8) 2 877 (76.0) 70–79 5 990 7 202 4 569 (76.3) 5 529 (76.8) 3 914 (65.3) 4 634 (64.3) 80– 2 694 4 688 1 317 (48.9) 1 971 (42.4) 1 055 (39.2) 1 596 (34.0) All 12 162 15 676 8 784 (72.2) 10 731 (68.5) 7 604 (62.5) 9 107 (58.1)

Table 6 Number of invited, attending and included men and women in the Tromsø 6 and HUNT 2 surveys combined. In the study sample, participants with a follow-up time below one year (425 participants) and with missing information regarding cause-specific mortality (5 participants) or questionnaire data concerning smoking, marital status or level of education (2 374 participants) were excluded.

(28)

3.4 Ethics

All participants gave written, informed consent upon inclusion in the various surveys and had the option of withdrawing from participation after inclusion. All surveys were approved by the regional boards of research ethics and the Data Inspectorate through the Norwegian Social Science Data Services.

4 Methods

4.1 Assessment of Nutritional status

4.1.1 Body Mass index (Paper l-IV)

Participants had their weight (kg) and height (cm) measured to the nearest decimal at the research centres. During these measurements, participants wore light clothing and did not wear shoes.

BMI was calculated as weight (kg) divided by height (m) squared (kg/m²).

4.1.2 Waist circumference (Paper IV)

Waist circumference (WC) was measured horizontally to the nearest centimetre at the height of the umbilicus using steel tape. WC was available for all HUNT-2 participants and for all Tromsø participants aged 65-74 as well as for a random sample of participants aged 75-84 years.

4.1.3 Malnutrition Universal Screening Tool (Paper ll and III)

The Malnutrition Universal Screening Tool (MUST) was originally developed by the British Society of Parenteral and Enteral Nutrition (www.bapen.org.uk). It includes a grading of both (1) BMI and (2) weight loss in three categories in addition to (3) an acute disease component (figure 3). In the questionnaire used in the Tromsø 6 survey, participants were asked about any

involuntary weight loss during the previous six months (and if so, how many kg). Weight loss

(29)

values were placed in the following groups: below 5%, between 5% and 10% or above 10% of body weight prior to weight loss. The MUST also includes an acute disease component

corresponding to no nutritional intake for >5 days, which normally necessitates hospitalisation [36]. As participation in the Tromsø study required the ability to independently visit a research centre, the acute diseases component was set to zero. The weight loss question was slightly modified to indicate a time span of the “last 6 months”, but this encompasses the time span of

“the past 3-6 months” in the original MUST. The use of the MUST as described in paper II and III has been confirmed by Professor Marinos Elia, who is in charge of the use of the tool on behalf of BAPEN (appendix B).

Add scores

Score=O Score=1 Score≥≥≥≥2 Low risk Medium Risk High Risk

Overall risk of malnutrition

If patient is acutely ill and there has been or is likely to be no nutritional intake for >5 days

Score2 Unplanned weight

loss in past 3-6 months Score

<5 % =0 5-10 % =1

>10 % =2 BMI(kg/m²) Score

>20 (>30 Obese) =0

18.5-20 =1

<18.5 =2

The "Malnutrition Universal Screening Tool" is reproduced here with the kind permission of BAPEN (British Association for Parenteral and Enteral Nutrition). For further information on 'MUST' and management guidelines, see www.bapen.org.uk.

Figure 3 The Malnutrition Universal Screening Tool (MUST) is composed of a BMI score, a weight-loss score and an acute illness component. The risk of malnutrition can be assessed based on the sum of these scores.

(30)

4.2 Data on cancer (Papers I and IV) and marital status (Papers I to IV)

Data concerning cancer history was obtained from the Norwegian Cancer Registry, which is based on mandatory registration. Data on marital status was obtained from the National Population Register.

4.3 Hand grip strength (Paper I)

Grip strength can be used as a measure of overall muscle strength [14] and has found to be an indirect measure of lean body mass [138]. In a representative subgroup of the Tromsø 4 population, grip strength of the non-dominant hand was registered in kilopascals (kPa), with measurements generated by manual compression of an air filled rubber bulb connected to a manometer. A measurement below the median value for each sex was defined as low grip strength.

4.4 Assessment of Mortality (Paper IV)

For the assessment of mortality, each participant included in paper IV (Tromsø 4 and HUNT 2 combined) were linked to information from the Norwegian Causes of Death Registry by a personal identification number to identify vital status (dead, alive or emigrated) during the follow-up period. Data concerning both total and cause-specific mortality were available. When we started on the analyses used in paper IV, data concerning cause specific mortality was available up to 31 December 2007, and this date was set as end date of follow up (figure 4).

Cause of death in Norway is routinely registered and coded by the Death Registry based on the International Classification of Diseases (ICD), and we noted the underlying cause of death. For deaths up to 1996, ICD version 9 was used and ICD version 10 was used for later years. We

(31)

applied the European short list for causes of death [42] to identify three main categories of causes of death; cardiovascular diseases (CVD, ICD-9: 390-459, ICD-10: I00-I99), respiratory diseases (ICD-9: 460-519, ICD-10: J00-J98) and cancer (ICD-9: 140-208, ICD-10: C00-C97)).

Figure 4 Timeline for the follow-up in paper IV (see 5.4 for details regarding the study).

4.5 Self-administrated Questionnaires (Paper I-IV)

In all three surveys included in this thesis, self-administrated questionnaires (appendix C) were used to obtain information concerning a wide range of diseases and symptoms, smoking habits, alcohol intake, social conditions, education, level of physical activity and other variables. In all surveys, participants were asked to fill in two different questionnaires. The first questionnaire was included with the invitation letter sent by mail to all participants. This questionnaire was intended to be completed at home and collected at the research centre. The second questionnaire was given to the participants upon admittance and was intended to be returned by mail in pre- stamped envelopes. In the HUNT 2 and Tromsø 4 surveys, the second questionnaire was given in different versions for those below and above the age of 70.

In paper I, we selected symptomatic medical conditions prevalent in the elderly population that may have a connection to either low weight or obesity according to the literature [55, 143]. The

(32)

considered conditions were mental distress, hip fracture, asthma or chronic bronchitis, stroke, angina pectoris or myocardial infarction and diabetes mellitus. Cancer was also included, but (as noted above) based on information from the Norwegian cancer registry. Chronic lung disease was defined as asthma or chronic bronchitis, and similarly, ischemic heart disease (IHD) was defined as myocardial infarction or angina pectoris (Paper I and IV).

4.6 Assessment of mental health: CONOR mental health index (Paper I)

In paper I (based on the Tromsø 4 survey), participants’ mental health was explored as one of several health conditions. This was done by means of an index based on seven questions concerning different dimensions of mental distress (CONOR Mental Health Index) [134]. The index was developed for use in a cohort including several health surveys in Norway; the COhort NORway (CONOR).This mental distress index was partly modified from the Hopkins Symptom Check List [27] and the General Health Questionnaire (GHQ) [53]. A cut-off value of 2.15 has been proposed to identify persons with significant mental distress [134].

4.7 Assessment of mental health: Symptoms Check List 10 (Paper II)

In paper II (the Tromsø 6 survey), mental health was assessed in greater detail by the Hopkins Symptoms Check List-10 (SCL-10). The SCL-10 has been widely used in epidemiological studies and is a self-administrated instrument that mainly explores symptoms of anxiety and depression [139]. The ten items of the SCL-10 were part of the questionnaire included in the invitation to the Tromsø 6 survey (appendix C).

(33)

The SCL-10 questions explored the presence and severity of the following ten symptoms during the preceding week: (1) “Sudden fear without apparent reason”, (2) “Afraid or worried”, (3)

“Faintness or dizziness”, (4) “Tense or upset”, (5) “Easily blaming yourself ”, (6)

“Sleeplessness”, (7) “Depressed or sad”, (8) “Feeling worthless”, (9) “Feeling that everything is a struggle”, and (10) “Feeling hopelessness with regard to the future”.

Each question was rated on a four-point scale ranging from 1 (not at all) to 4 (extremely).

According to a procedure suggested by Strand et al.[139], missing values were replaced by the sample mean value for each item, but questionnaires with three or more missing values were excluded from the analyses. The SCL-10 total score was calculated by dividing the total score by the total number of items (i.e., ten). A higher score indicated more symptoms. We found an acceptable degree of internal consistency for this scale in the study sample in paper II (Cronbach’s alpha = 0.84).

An SCL-10 score of 1.85 has been proposed as the cut-off for predicting diagnosed mental disorders [139], and score values of ≥1.85 in the current study were referred to as significant symptoms. To assess the impact of score values below this cut-off, we subdivided the SCL-10 scores in paper II between 1.01 and 1.84 into a lower score category (SCL-10 score 1.01 to 1.39) referred to as some symptoms, and a higher score category (SCL-10 score 1.40 to 1.84) referred to as sub-threshold symptoms. Individuals with no symptoms (SCL-10 score 1.0) were used for reference in the analyses.

(34)

4.8 Assessment of Health Related Quality of Life: EQ-5D (Paper III)

HRQoL (Paper III) was assessed by the EuroQol questionnaire (EQ-5D), which has been developed by a multidisciplinary group of European researchers. This is a standardised non- disease specific instrument consisting of two parts; the EQ-5D descriptive system and the EQ visual analogue scale (EQ VAS) [148]. The EQ-5D instrument has been evaluated in a number of studies and has been validated in elderly populations [17, 21].

The EQ-5D describes health in generic terms using five specific dimensions that are important for elderly individuals: mobility, self-care, usual activities, pain/discomfort and

anxiety/depression. Each dimension is divided into three levels of severity (no problems, some problems or extreme problems). In the sample described in paper III, few participants reported problems at the most severe level. According to EQ-5D user guide [41], this category was combined with the category of individuals reporting some problems (second level) in the analyses of the various EQ-5D dimensions. The EQ-5D instrument is designed for self- completion and was included as part of the self-administrated questionnaire in the Tromsø 6 survey (appendix C). A single summary EQ-5D index with a maximum score of 1 is obtained by applying a scoring algorithm that assigns weights to each of the possible combinations of health as described by the three levels within each of the five dimensions. We applied the most widely used scoring algorithm, referred to as the UK time-trade-off tariff [32]. Subjects with missing values to any of the five dimensions were excluded from the analyses.

(35)

In addition to this indirect health index assigned through a descriptive system, a direct method asks subjects to rate their health on a visual analogue scale (VAS) with a maximum score 100.

The endpoints are labelled “Worst imaginable health state” and “Best imaginable health state”.

4.9 Statistical methods

Statistical analyses were performed using the Statistical Package for Social Science (SPSS), versions 15.0 and 17.0 (SPSS Inc, Chicago, Illinois, USA). Means, medians, and proportions (%) were used to describe both baseline characteristics and outcomes, when appropriate. Student’s t- tests were used to analyse differences between mean scores and chi-square test for differences in proportions. Mann-Whitney U test were applied when appropriate. Two sided P-values <0.05 were considered statistically significant. Odds ratio (OR) and hazard ratio (HR) point estimates were reported with 95% confidence intervals. Adjustments were performed for potential confounding variables (see section 6.4, page 57 regarding what was considered confounding).

Most analyses were stratified by sex. Logistic regression models were used in papers I-III to assess associations between BMI and/or risk category of malnutrition, and dichotomised chronic diseases, social and lifestyle variables (paper I), SCL-10 score category (paper II), or EQ-5D dimensions (paper III). In the analyses of specific variables, cases with missing values were not included.

In paper III, analysis of covariance was used to obtain age-adjusted mean values for the EQ-5D score. The importance of the differences in EQ-5D scores between malnutrition risk groups was examined by calculating their effect size as the mean difference divided by the standard deviation (SD) of the control group [43]. We evaluated the detected differences against the criteria

(36)

introduced by Cohen [23] using the standard deviation (SD) of the low-risk category of

malnutrition. Effect size values of 0.2 to <0.5, 0.5 to < 0.8, and ≥0.8 were characterised as small, medium and large differences, respectively.

Mortality analyses in paper IV were performed using a Cox proportional hazards regression model. We assessed the proportional hazards assumption of a constant hazard ratio over time by inspecting the log-log survival curves for the various BMI categories.

For logistic regression analyses of the various BMI categories (paper I), we used the BMI

category 22.5-24.9 kg/m² as a reference because this is the upper normal BMI category according to WHO [164]. In papers II-IV, the BMI category 25-27.4 kg/m² was used as reference, as this was the category both with the most participants and with the highest number of deaths (paper IV).

For further details, see the description of data analyses in the various papers.

(37)

5 Summaries of papers and main results

In papers I-IV, we explored the relationships between BMI, the risk of malnutrition and morbidity, HRQoL and mortality (figure 5).

Figure 5 The aspects of health described in papers I through IV.

Paper I identifies the potential risk factors, chronic conditions and diseases associated with different BMI categories. Paper II explores mental health using the SCL-10 score. Another important dimension of life for elderly individuals is HRQoL, which is explored in paper III.

Participants examined in papers II and III were classified both according to their BMI and risk of malnutrition. In paper IV, the relationships between the various BMI categories and mortality, the ultimate endpoint, were explored. For more details, see the sections 5.1 to 5.4.

(38)

5.1 Paper I

Body mass index and disease burden in elderly men and women: The Tromsø Study Jan-Magnus Kvamme, Tom Wilsgaard, Jon Florholmen, Bjarne K Jacobsen.

Eur J Epidemiol, 2010: 25(3); 183-193. Epub 2010 Jan 20.

Background: Chronic health problems may be related to body mass index (BMI), but this has been best documented in overweight and obese adults. The primary objective of this study was to identify factors associated with different categories of BMI in community-living elderly men and women, also including the lower categories of BMI.

Methods: In a cross-sectional population survey from the municipality of Tromsø, Norway, we analysed associations between BMI and a wide range of chronic disease conditions, lifestyle and socioeconomic factors. BMI (kg/m²) was categorised into six groups (<20 kg/m², 20.0-22.4 kg/m², 22.5-24.9 kg/m², 25.0-27.4 kg/m², 27.5-29.9 kg/m², ≥30.0 kg/m²). The study included 4 259 non-institutionalised men and women aged 65 years and older.

Results: The overall proportion with BMI <20 kg/m² (low weight) was 5.1%, BMI between 25.0 and 29.9 kg/m² (overweight) 42.1% and BMI ≥30.0 kg/m² (obesity) 16.9%. Obesity was more common in women than in men (21.8% vs. 10.4%). Current smoking, mental distress and hip fractures were more prevalent in the lower BMI categories in both sexes. Asthma or chronic bronchitis and low physical activity exhibited a U-shaped relation to BMI (p-value for a second order term <0.05). Neither single marital status, difficult economy nor lower education was related to underweight, whereas lower education and a difficult economy were related to obesity.

Reduced muscle strength was correlated to low weight. Alcohol intake was less frequent in the higher categories of BMI. Diabetes mellitus and ischemic heart disease (IHD) were associated

(39)

only with higher BMI categories. Diabetes mellitus was associated with obesity and IHD was associated with overweight and obesity.

Conclusion: These results demonstrate that both low and high BMI are associated with a wide range of prevalent conditions and diseases in community-living elderly individuals. For the clinician, the findings emphasise the importance of nutritional assessment as part of the medical evaluation of elderly patients.

5.2 Paper II

Risk of malnutrition and mental health symptoms in community-living elderly men and women: The Tromsø Study

Jan-Magnus Kvamme, Ole Grønli,Jon Florholmen, Bjarne K Jacobsen. Submitted

Background: Little research has been done to examine the relationship between the risk of malnutrition and mental health in community-living elderly individuals. In the present study, we aimed to assess the associations between mental health (particularly anxiety and depression) and both the risk of malnutrition and body mass index (BMI, kg/m²) in a large sample of elderly men and women from Tromsø, Norway.

Methods: In a cross-sectional survey of 1 558 men and 1 553 women aged 65 to 87 years, the risk of malnutrition was assessed by the Malnutrition Universal Screening Tool (MUST), and mental health was measured using the Symptoms Check List 10 (SCL-10). BMI was categorised into six groups (<20.0 kg/m², 20.0-22.4 kg/m², 22.5-24.9 kg/m², 25.0-27.4 kg/m², 27.5-29.9 kg/m² , ≥30.0 kg/m² ).

Results: The risk of malnutrition (combining medium and high risk) was found in 5.6% of men and 8.6% of women. Significant mental health symptoms were reported by 3.9% of men and 9.1% of women. In a model adjusted for age, marital status, smoking and education, significant

(40)

mental health symptoms (SCL-10 score ≥1.85) were positively associated with the risk of malnutrition (odds ratio 3.9 [95% CI 1.7-8.6] in men and 2.5 [95% CI 1.3-4.9] in women), the association was positive also for subthreshold mental health symptoms. For individuals with BMI

<20.0 kg/m², the adjusted odds ratio for significant mental health symptoms was 2.0 [95% CI 1.0-4.0]. No significant increase in significant mental health symptoms was found in obese individuals.

Conclusions: Impaired mental health was strongly associated with the risk of malnutrition in community-living elderly men and women, and this association was also significant for subthreshold mental health symptoms.

5.3 Paper III

Risk of malnutrition and health-related quality of life in community-living elderly men and women: The Tromsø Study

Jan-Magnus Kvamme, Jan Abel Olsen, Jon Florholmen, Bjarne K Jacobsen.

Qual Life Res. 2011 May;20(4):575-582. Epub 2010 Nov 13.

Background: Health-related quality of life (HRQoL) has received increased attention in previous decades as a measure for both comparing health statuses across different patient groups and for measuring health outcomes. However, larger community-based studies evaluating HRQoL in elderly individuals at risk of malnutrition are lacking. In the present study, we aimed to explore the association between risk of malnutrition and HRQoL in community living elderly men and women. The association between body mass index (BMI) and HRQoL was also explored.

(41)

Methods: In a cross-sectional population survey including 1 654¹ men and 1 632 women aged ≥ 65 years from the municipality of Tromsø, Norway, we assessed HRQoL by using the EuroQol (EQ-5D) instrument in three risk groups of malnutrition and in different categories of BMI. The Malnutrition Universal Screening Tool (MUST) was used to evaluate the risk of malnutrition.

Results: More women (9.4%) than men (5.5%) were at risk of malnutrition (medium- and high- risk combined). HRQoL was lower in women than in men when assessed by the EQ-5D index and the EQ VAS score. We found a significant reduction in HRQoL with increasing risk of malnutrition, and this was more pronounced in men than in women. The relation between BMI and HRQoL was dome shaped, with the highest HRQoL scores being observed in the 25-27.5 kg/m² BMI category although the differences were small between the middle BMI categories.

Conclusions: HRQoL was significantly reduced in elderly men and women at risk of malnutrition. The highest HRQoL was seen in moderately overweight individuals.

5.4 Paper IV

Body mass index and mortality in elderly men and women: The Tromsø and HUNT studies Jan-Magnus Kvamme, Jostein Holmen, Tom Wilsgaard, Jon Florholmen, Kristian Midthjell, Bjarne K Jacobsen. J Epidemiol Community Health. Published online February 14, 2011 Background: The impact of body mass index (BMI, kg/m²) and waist circumference (WC) on mortality in elderly individuals is controversial, and previous research has largely focused on obesity.

Methods: With special attention to the lower BMI categories, associations between BMI and both total and cause-specific mortality were explored in 7 604 men and 9 107 women aged ≥65 years who participated in the Tromsø 4 survey (1994-95) or the North-Trøndelag Health Study

1 In the abstract of the paper III, the numbers of participating men and women had been interchanged. The correct numbers are 1654 men and 1632 women as written in the table 1 in the main part of the paper.

(42)

(HUNT 2 survey) (1995-97). A Cox proportional hazards model adjusted for age, marital status, education, study site and smoking, was used to estimate hazard ratios for mortality in different BMI categories using the BMI range of 25-27.5 kg/m² as a reference. The impact of each 2.5 kg/m² difference in BMI on mortality (later in this abstract denoted as change in BMI) in individuals with BMI <25.0 kg/m²and BMI ≥25.0 kg/m²was also explored. Furthermore, relationships between WC and mortality were assessed.

Results: We identified 7 474 deaths during a mean follow-up period of 9.3 years until

31.12.2007. The lowest mortality was found in the BMI range 25-29.9kg/m²and 25-32.4 kg/m² in men and women, respectively. Mortality was increased in all BMI categories below 25 kg/m² and was moderately increased in obese individuals. U-shaped relationships were also found between WC and total mortality. When modelling BMI as a continuous variable, we found a 20%

increase in mortality per 2.5 kg/m² decrease in BMI in the lower BMI range (<25 kg/m²). In the upper BMI range (≥25 kg/m²), we found a 7 to 9% increase in mortality per 2.5 kg/m² increase in BMI. About 40% of the excess mortality in the lower BMI range in men was explained by mortality from respiratory diseases.

Conclusions: BMI below 25 kg/m² was in elderly men and women associated with increased mortality. A modest increase in mortality was found with increasing BMI among obese men and women. Overweight individuals (BMI 25-29.9 kg/m²in men, 25-32.4 kg/m² in women) had the lowest mortality.

(43)

6 General Discussion – Methodology

6.1 Selection of populations and study design

Each study included in this thesis is population-based, which is important for the generalisability of the results. We aimed at describing populations of community-living elderly men and women as distinct from elderly hospital in-patients. Nursing home patients were invited, but attended to a very small degree (8 out of 4 017 participants in the Tromsø 6 survey, and 45 out of 15 250 in the HUNT 2 survey). Consequently, the participants in the included studies may be regarded as community-living individuals.

In papers I-III, we explore associations in a cross-sectional design, which restricts the possibility of making cause-effect conclusions. In paper IV, we used a prospective cohort design, which makes it possible to derive causality-based conclusions.

6.2 Validity

The aim of epidemiological research (e.g., the studies included in this thesis) is to obtain correct and precise results that can be generalised to other populations. The validity of a study refers the extent to which this aim is fulfilled. Internal validity concerns the degree to which it is possible to draw conclusions concerning the study population. External validity concerns the degree to which it is possible to apply the results to other populations (i.e., generalisability of the results) [90, 119] (figure 6). Internal validity is regarded as a prerequisite for external validity.

(44)

6.3 Internal validity and bias

Bias is systematic error that tends to produce results that depart systematically from the true values [71, 93]. Violations of internal validity can be classified into one of the categories selection bias, information bias or confounding (figure 6) [119].

Figure 6 Internal and external validity with examples from the thesis. Internal validity is regarded as a prerequisite for external validity (curved dotted arrow).

6.3.1 Selection bias

Selection bias may arise when the subjects included in the study sample differ from the source population in a way that affects the conclusions [90]. This type of bias arises from participant selection procedures or from factors that influence participation in the study. For selection bias to occur, the association between the independent and dependent variables must be different

(45)

between study participants and non-participants [123]. The potential for selection bias is

somewhat limited by the attendance rates in the included surveys, which were between 65% and 70% (See tables 4 to 6). This may be regarded as relatively high for population-based

epidemiological studies. However, there were non-responders for several of the variables, which reduced the size of the study sample by an additional 8-20 percentage points (table 5 and 6).

A precise measurement of the effect of selection bias is not possible, as this would require

information regarding the exposure and outcome status for both participants and non-participants.

[15]. The only information we have about the non-participants in the Tromsø and HUNT surveys is age and sex. The mean age of the participants included in the study samples was generally lower than for non-attending individuals. In all papers, there was a response rate below 50% for the highest age group (≥80 years).

However, a small study of subjects who selected not to participate in the HUNT 2 survey, was performed shortly after completion of the fieldwork [68]. Among individuals aged ≥70 years reached by phone, immobility due to disease and sufficient follow-up by the family doctor were reported as important reasons for not participating. In the Tromsø 4 and 6 surveys, immobility was likely more common in non-participating elderly individuals, making it difficult to visit the research centres. Moreover, both decline in cognitive function and dementia are conditions likely to be more prevalent in non-attending individuals and in individuals with missing data on the self-administrated questionnaires. Dementia generally affects 10-20% of the population aged >80 years [37] and is associated with the risk of malnutrition [70].We do not have data on the

prevalence of dementia in the elderly in the Tromsø and HUNT surveys.

Body mass index and risk of malnutrition in community-living elderly men and women: relationships with morbidity, mortality and health- related quality of life. The Tromsø and HUNT studies

FACULTY OF HEALTH SCIENCES

DEPARTMENT OF COMMUNITY MEDICINE